Since deep learning took off in 2012, Face recognition has become an exciting market with multiple companies offering a diverse array of solutions. Integrating face recognition capabilities into your project has never been easier. If you are not sure about what face recognition SDK/API to choose, this article will help you.
There are primarily two types of solutions, Face Recognition SDKs and Face Recognition APIs.
APIs are usually cloud hosted and are available on a subscription basis. There are limits on the number of queries per second or per month and on the face database storage capacity. Since they are cloud hosted, these APIs are accessible from any platform/device and extremely easy to integrate in your existing codebase. It is also easy to switch services if desired. The deep learning algorithms run on the provider's infrastructure and this alleviates a major burden of managing GPU compute resources. However, APIs have a significant drawback - they cannot be used for real time applications. On the one hand, they will become very expensive and two, uploading 30 images per second over the internet is not feasible in any realistic sense. They also require a reliable network connection. Apart from these, one should also pay attention to the pricing of the vendor. Some dynamic pricing plans have can cause monthly bills to shoot-up exponentially!
SDKs, on the other hand, are either available as pre-built libraries or provided along with source code for integration in the application. They work completely offline and upon build/compilation, the program becomes a single unit containing the face detection/recognition algorithm. This also makes it feasible for real-time applications. There are no limits on the number of queries or storage capacity. However, the infrastructure to deploy this application must be managed by you. They are also quite expensive to license compared to cloud APIs.
The table below summarizes the above
Face Recognition API |
Face Recognition SDK |
Limitations
|
Limitations
|
Here is a summary of the different competitive Face Recognition APIs currently available in the market in no particular order
Service\Feature |
Detection |
Landmarks |
Recognition |
Age/Gender |
Emotion |
Head Pose |
Other Classifiers |
BetaFace API |
Yes | 123 | Yes | Yes | Smile/Neutral | No | Facial hair, Glasses |
Face++ |
Yes | 83 or 106 | Yes | Yes | 7 classes | Yes | Face quality Blur Eye status Ethnicity Beauty Mouth status Eye gaze Skin |
Amazon AWS Rekognition |
Yes | 23 | Yes | Yes | 3 Classes + Smile intensity |
Yes | Beard Mustache Sunglasses EyesOpen Mouth Open |
Microsoft |
Yes | 27 | Yes | Yes | Smile Intensity | Yes | Facial hair, glasses |
Google Cloud Platform |
Yes | No | No | No | Smile/Neutral | No | Headwear |
IBM Watson Face |
Yes | No | Yes | Yes | No | No | None |
Animetrics FaceR API |
Yes | 24 | Yes | Yes | No | Yes | None |
Kairos API |
Yes | 49 | Yes | Yes | 6 classes | Yes | Ethnicity Eye status |
Meerkat FR |
Yes | 68 | Yes | No | No | Yes | None |
Here is a summary of the different competitive Face Recognition SDKs currently available in the market
Service\Feature |
Platforms |
Detection |
Landmarks |
Recognition |
Age/Gender |
Emotion |
Head Pose |
Other Classifiers |
BetaFace SDK |
Win/Lin | Yes | 123 | Yes | Yes | Smile/Neutral | No | Facial hair, Glasses |
Cognitec VACs |
Win/Lin/ARM | Yes | unknown | Yes | Yes | No | Yes | Glasses Eye status Lighting, Color Image and face geometry Full-frontal image check as required for ePassport |
Luxand FaceSDK |
Win/Lin/ARM | Yes | 70 | Yes | Yes | Smile/Neutral | Yes | Eye status |
NeuroTech VeriLook SDK |
Win/Lin/ARM | Yes | No | Yes | No | No | No | None |
SightCorp Insight SDK |
Win/Lin/ARM | Yes | unknown | No | Yes | 7 classes | Yes | None |
Animetrics FaceR API |
Win/Lin | Yes | 24 | Yes | Yes | No | Yes | None |
Visage SDK |
Win/Lin/ARM | Yes | 68 | Yes | Yes | 6 classes | Yes | Eye status, Gaze tracking |
Affectiva SDK |
Win/Lin/ARM | Yes | No | False | Yes | 7 classes | Yes | Ethnicity |
Deepsight SDK |
Win/Lin | Yes | 68 | Yes | Yes | soon | soon | None |
